idea-mining — a Claude skill for structured product brainstorming

A reusable Claude skill that turns single-shot brainstorming into a structured 100-idea mining loop. Bilingual (English + French).

What it does

When you ask Claude for ideas — for a product, feature, optimization, business strategy — the default response is 5-10 predictable suggestions because the model converges on the modal answer. This skill changes the loop:

1. Captures context — what you're building, for whom, what metric matters, what the baseline is
2. Generates a structured mining prompt — 10 sub-agents in parallel, each constrained to a distinct axis of divergence
3. Produces ~100 raw ideas — organized, tagged with impact and effort estimates
4. Helps you filter — surfaces the 3-5 genuinely non-obvious candidates worth simulating

The method comes from a publicly shared workflow used by early Claude Code adopters (the "Clarbo 100-ideas loop"). This skill packages it as a reusable Claude asset with bilingual support and domain-specific axis libraries.

When to use it

Trigger phrases (the skill listens for these and similar in French and English):

• "I need ideas for [problem]" / "J'ai besoin d'idées pour [problème]"
• "How can we improve [metric]?" / "Comment améliorer [métrique] ?"
• "What are all the ways to [goal]?" / "Quelles sont toutes les façons de [objectif] ?"
• "Brainstorm with me on [topic]" / "Brainstorming sur [sujet]"
• "Help me find features for [product]" / "Aide-moi à trouver des features pour [produit]"

The skill works best on strategic questions with multiple valid directions. It's not useful for single-answer factual questions or tasks where you already know what to build.

Installation

Claude.ai (web/desktop)

1. Download the latest .skill file from the Releases tab
2. In Claude.ai, go to Settings → Capabilities → Skills
3. Upload the .skill file

Claude Code

1. Clone or download this repo
2. Copy the idea-mining/ folder into your project's .claude/skills/ directory:

   cp -r idea-mining ~/.claude/skills/

3. The skill will load automatically on next Claude Code session

Manual / other clients

Copy the idea-mining/ folder into wherever your Claude client loads skills from. The skill is pure markdown — no scripts, no dependencies.

Repository structure

idea-mining/
├── SKILL.md                              # main skill definition
└── references/
    ├── prompt-template.md                # mining prompt template (EN + FR) + worked example
    ├── divergence-axes.md                # library of 30+ axes by product type
    ├── filtering-heuristics.md           # how to shortlist after the mining round
    └── worked-examples.md                # three full end-to-end examples

How it works under the hood

The skill encodes a method, not just a prompt. When triggered, Claude:

1. Reads the user's product context (asks 2-3 questions max if anything's missing)
2. Picks 10 axes of divergence from the library, adapted to the user's domain
3. Constructs the mining prompt with explicit anti-self-censorship and anti-duplicate constraints
4. Runs the mining inline (or hands off the prompt for separate execution)
5. Helps the user filter: flags duplicates, surfaces non-obvious winners, identifies testable assumptions
6. Suggests next steps (simulation, feature spec, deeper mining round)

The full method is documented in idea-mining/SKILL.md and the reference files.

Examples included

The references/worked-examples.md file walks through three complete sessions:

1. Maritime routing (deeptech, French) — differentiating a routing engine for mid-market shipowners
2. Creator economy tool (consumer, English) — improving install-to-WAU for a YouTuber Chrome extension
3. B2B sales platform (SaaS, bilingual) — increasing lead-to-quote conversion for a construction-sector CRM

Each example shows the captured context, the axes chosen, the winning shortlist, and the rejected ideas with reasons.

License

Free to use, modify, redistribute. The Clarbo method itself is publicly shared; this skill packages it.

Attribution to the original Clarbo founders and to this repo is appreciated but not required.

Feedback and contributions

Open an issue if:

• A trigger phrase doesn't reliably activate the skill for you
• A domain you work in needs custom axes not yet in divergence-axes.md
• A worked example would help in your vertical

Pull requests welcome, especially for new axis libraries by domain and additional worked examples.

---

Built and shared by Matthieu / OceanData Consulting. If you're working on maritime routing, AIS ML, or cloud-native scientific data, the rest of my work is over there.